R Notebook

sqrt(-sum(log(fitCCstderr)))

[1] 5.41743

sqrt(-sum(log(fitMICEstderr)))

[1] 63.71021

sqrt(-sum(log(fitMRstderr)))

[1] 63.59802

sqrt(-sum(log(fitMIDASstderr)))

[1] 63.86592

library(plotly)
p <- plot_ly(
  x = c("Complete case", "MIC", "missRanger", "MIDAS"),
  y = c(-sum(log(fitCCstderr)), -sum(log(fitMICEstderr)), -sum(log(fitMRstderr)), -sum(log(fitMIDASstderr))),
  name = "Sum of logged Std Errors",
  type = "bar"
)
p

cbind(MICEloss = MICEmeanloss, MRloss = MRmeanloss, MIDASloss = MIDASmeanloss, MIDASVAEloss = MIDASVAEmeanloss, CCLoss=CCmeanloss, adjustedCCLoss = adjustedCCmeanloss)

      MICEloss    MRloss MIDASloss MIDASVAEloss    CCLoss adjustedCCLoss
[1,] 0.2668602 0.2673009 0.2668431    0.2672094 0.3712361      0.3044155

#dfp = cbind(Complete = CClossvector, MICE = MICElossvector, missRanger = MRlossvector, MIDAS = MIDASlossvector)
#plot_ly(midwest, x = ~percollege, color = ~state, type = "box")
#plot_ly(data.frame(CClossvector), type = "box")
method = c(rep("Complete",length(CClossvector)),rep("MICE",length(MICElossvector)),rep("missRanger",length(MRlossvector))
           ,rep("MIDAS",length(MIDASlossvector)))
value = c(CClossvector, MICElossvector, MRlossvector, MIDASlossvector)
data=data.frame(method,value)
qplot( x=method , y=value , data=data , geom=c("boxplot") , fill=method)

cmmice

Confusion Matrix and Statistics

          Reference
Prediction    0    1
         0 9449  972
         1  147  179
                                          
               Accuracy : 0.8959          
                 95% CI : (0.8899, 0.9016)
    No Information Rate : 0.8929          
    P-Value [Acc > NIR] : 0.1629          
                                          
                  Kappa : 0.2048          
 Mcnemar's Test P-Value : <2e-16          
                                          
            Sensitivity : 0.15552         
            Specificity : 0.98468         
         Pos Pred Value : 0.54908         
         Neg Pred Value : 0.90673         
             Prevalence : 0.10710         
         Detection Rate : 0.01666         
   Detection Prevalence : 0.03033         
      Balanced Accuracy : 0.57010         
                                          
       'Positive' Class : 1               
                                          

cmmice = confusionMatrix(as.factor( round(MICEpreds)),as.factor(outcome), positive='1')
cmmr = confusionMatrix(as.factor( round(MRpreds)),as.factor(outcome), positive='1')
cmmidas = confusionMatrix(as.factor( round(MIDASpreds)),as.factor(outcome), positive='1')
cmcc = confusionMatrix(as.factor( round(CCpreds)),as.factor(CCoutcome), positive='1')
# cmmice$byClass["Pos Pred Value"]
#   overall["Accuracy"]
method = c("Complete case", "MICE", "missRanger", "MIDAS")
metric = c("Accuracy", "Brier", "Pos Pred Value", "Neg Pred Value", "Mean Loss", "AUC")
accuracy = c(cmcc$overall["Accuracy"],cmmice$overall["Accuracy"],cmmr$overall["Accuracy"],cmmidas$overall["Accuracy"])
accuracy = accuracy/accuracy[1] * 100
Brier = c(CCbrier, MICEbrier, MRbrier, MIDASbrier)
Brier = Brier/Brier[1] * 100
posPredValue = c(cmcc$byClass["Pos Pred Value"],cmmice$byClass["Pos Pred Value"],cmmr$byClass["Pos Pred Value"],cmmidas$byClass["Pos Pred Value"])
posPredValue =   posPredValue/posPredValue[1] * 100
negPredValue = c(cmcc$byClass["Neg Pred Value"],cmmice$byClass["Neg Pred Value"],cmmr$byClass["Neg Pred Value"],cmmidas$byClass["Neg Pred Value"])
negPredValue = negPredValue/negPredValue[1] * 100
meanloss = c(CCmeanloss, MICEmeanloss, MRmeanloss, MIDASmeanloss)
meanloss = meanloss/meanloss[1] * 100
auc = c(CCci[2], MICEci[2],MRci[2], MIDASci[2])
auc = auc/auc[1] * 100
# CCci[1]
# CCci[2]
# CCci[3]
# CCci
#summarydf = data.frame(method, accuracy, accuracyN, precision, precisionN, negpredvalue, negpredvalueN, meanloss, meanlossN, auc,
summarydf = data.frame(method, accuracy, precision, negpredvalue, meanloss, auc)

# color with RcolorBrewer
ggplot(summarydf, aes(fill=method, y=value, x=specie)) + 
    geom_bar( stat="identity", position="fill") +    
    scale_fill_brewer(palette = "Set1")
 
# Faceting
ggplot(data, aes(y=value, x=specie, color=specie, fill=specie)) + 
    geom_bar( stat="identity") +    
    facet_wrap(~condition)

library(reshape)
library(data.table)
sdf2 = data.frame(t(summarydf))
sdf2 

names(sdf2) = summarydf$method
setDT(sdf2, keep.rownames = TRUE)[]

colnames(sdf2)[1] <- "Metric"
#sdf3 = sdf2[c(3,5,7,9, 11),]
sdf3 = sdf2[-1]
sdf3$"Complete case"=round(as.numeric(as.character(sdf3$"Complete case")),1)
sdf3$"MICE"=round(as.numeric(as.character(sdf3$"MICE")),1)
sdf3$"missRanger"=round(as.numeric(as.character(sdf3$"missRanger")),1)
sdf3$"MIDAS"=round(as.numeric(as.character(sdf3$"MIDAS")),1)
sdf3

library(reshape2)
# reshape your data into long format
sdf3l <- melt(sdf3, id=c("Metric"))
sdf3l

#create color palette:
library(RColorBrewer)
coul = brewer.pal(4, "Dark2") 
#values = c("red","blue","green","yellow"), 
# make the plot
ggplot(sdf3l) +
  geom_bar(aes(x = Metric, y = value, fill = variable), 
           stat="identity", position = "dodge", width = 0.7) +
  scale_fill_manual("Method\n", values = coul, 
                    labels = c("Complete Case", "MICE", "missRanger", "MIDAS")) +
  labs(x="",y="Relative Performance %\n") +
  ylim(c(0,120)) +
  theme_bw(base_size = 14) + 
  theme(axis.text.x = element_text(angle = 60, hjust = 1))